Introduction

Here I develop statistical models to quantify the effect of forestry disturbance on caribou distribution. The intent is to use the models to indicate the influence of current and potential future forestry development on caribou use of habitat. Tne model will be implemented as a module (i.e., rsfCLUS) in the caribou and land use simulator (CLUS). Models were fit for the four caribou designatable units (DUs) in British Columbia (BC): DU6, boreal; DU7, northern mountain; DU8, central mountain; and DU9, southern mountain, as defined by the Committee on the Status of Endangered Wildlife in Canada (COSEWIC). DUs were modeled independently, as they are intended to represent groups of caribou with unique ecological characteristics, and potentially different responses to forestry disturbance.

I use a resource selection function (RSF) framework to develop the statistical models ( Boyce et al. 1999, Johnson et al. 2006). This framework uses a logit model to calculate the relationship between caribou occurence and habitat ‘resources’ measured using spatial data. Here I am only considering the effect of forestry disturbance ‘resources’, i.e., cutblocks and roads, on caribou occurrence. This approach simplifies and focuses the RSF onto modeling a statistical relationship between caribou distribution and forestry disturbance that is generalizable across a caribou DU, and is not designed to accurately identify other habitat features that influence caribou distribution. The theoretical basis for this approach is that forestry disturbance has a significant influence on caribou distribution, regardless of other habitat features. Indeed, cutblocks and linear features from forestry have routinely been shown to influence caribou distribution (e.g., Courbin et al. 2009, DeCesare et al. 2012, Mumma et al. 2018) and survival ( Wittmer et al. 2007, Johnson et al. 2015).

I fit RSF models with functional reponses by caribou to cutblocks and roads. Individual animals, including caribou, likely respond to disturbances differently (i.e., avoid or use roads or cutblocks) depending on the amount of disturbance occuring around them (i.e., the degree that their home range is fragmented by roads or cutblocks). Functional response models in RSFs empirically account for this theoretical response by including an interaction term between the disturbance feature measured at a location and the availability of that feature measured in the surronding environment ( Matthiopoulos et al. 2011). I hypothesized that caribou would be more likely to avoid cutblocks and roads as the amount of these features increased in their home ranges, as patches of habitat far from disturbances would become increasingly rare, and thus increasingly important to caribou for avoiding disturbance. I fit and evaluated RSF models with functional responses to cutblocks and roads in each DU and included them in the model if they were evidently influencing caribou occurrence.

The caribou-forestry disturbance models (CFDMs) created here can be used as a spatial indicator of current and simulated future effects of forestry activity on caribou habitat quality. For example, the sum of CFDM outputs can be calculated within caribou herd ranges or critical habitat types, and changes in the sum value over time can be used to indicate changes in forestry disturbance influencing caribou habitat quality.

Methods

I calculated the CFDMs using generalized linear mixed models (GLMMs) with a logit link, using the glmer function from the lme4 package in program R. I fit the model using caribou location data collected using telemetry collars placed on a sample of caribou (n = 468 animals, 507,987 locations) from across BC, between 2008 to 2018. These data were compiled from the government of BC’s Species Inventory (SPI) database, BC Oil and Gas Research and Innovation Society (OGRIS) data, and data from regional wildlife biologists in the BC government that was not yet entered into these databases. These data specified the locations on the landscape that were ‘used’ by caribou (sensu Boyce et al. 1999), i.e., the 1’s in the binomial distribution of the GLMMs dependent variable. I had no capture data for these animals, but it is likely the majority of the animals sampled were adult females, as the focus of caribou monitoring in BC is on measuring survival of adult females and the recruitment of their calves. Data was collected using a wide variety of telemetry collar technologies (e.g., GPS, ARGOS) and suppliers (e.g., Lotek, ATS, Telonics), and with varying location fix rates (e.g., 2 hours to 12 hours). All data were collected under provincial government permits and following provincial animal care and handling protocols.

I sampled locations ‘available’ to caribou, i.e., the 0’s in the binomial distribution of the GLMMs dependent variable, by calculating the seasonal home ranges of individual animals and randomly sampling locations within these areas. I calculated home ranges using the kernelUD function in the adehabitatHR ( Calenge et al. 2011) package of Program R. I varied the smoothing parameter (h) in the bivariate normal kernel algorithm and visually assesed the fit of 95% kernel polygons to the telemetry data. I selected the polygons with an h = 500, as they appeared to best fit the data ( Hemson et al. 2005, Calenge et al. 2011), i.e., that reasonably formed contiguous habitat areas (i.e., connected polygons) without including large extents greater than 1km from telemetry locations (Fig. 1). Animals with less than 50 locations in a season were removed from the analysis, as this is considered the minimum number of locations needed to calculate a home range using a kernel density algorithm ( Seaman et al. 1999, Kernohan et al. 2001). Seasons were defined as: summer (15-May to 31-Oct), early winter (01-Nov to 15-Jan) and late winter (16-Jan to 14-May), based on a review of the literature on variation in the seasonal distribution of caribou in British Columbia. Within each seasonal home range, I randomly sampled 1,000 locations.

Figure 1. Example of different smoothing factors (h) used to calculate caribou home ranges with a kernel density algorithm.

Figure 1. Example of different smoothing factors (h) used to calculate caribou home ranges with a kernel density algorithm.

I estimated the distance of ‘used’ and ‘available’ caribou locations to the nearest resource road and cutblock. Cutblock locations were estimated using the consolidated cutblocks data compiled by the BC government. Road locations were estimated using data compiled to support the British Columbia Cumulative Effects Framework. This data is not accessible to the public via a webpage, but it is not private or proprietary, and thus can be requested from the cumulative effects program. The roads data classifies roads into various types. Our focus was on roads developed for resource extraction (particularly forestry), thus, we classified all unpaved road types as ‘resource roads’ and estimated caribou distance to these types. Cutblock and roads data were rasterized at a 1ha resolution, and the Euclidean distance of each raster pixel to a cutblock or resource road in the pixel was calculated in ArcGIS and attributed to corresponding ‘used’ and ‘available’ caribou locations.

I fit GLMMs with distance to cutblock and distance to road covariates as fixed effects and random effects at the individual animal level. The random effects accounted for individual animal variation in response to cutblocks and roads in the model, so that the fixed effects coefficients can be considered as ‘population‐level’ effects (i.e.,the effect on an ‘average’ animal) of cutblocks and roads on caribou distribution ( Gillies et al. 2006). Similarly, I also examined the data for variability in herd-level response to cutblocks and roads, assuming animals from the same herd had similar reponses to cutblocks and roads, and fit GLMMs with random effects for cutblocks and roads at the herd level to account for that effect.

I fit models with and without functional responses to roads and cutblocks and visualy assessed whether a functional response was evident, i.e., caribou use of cutblocks and roads varied as the average distance to cutblocks and roads changed within their home ranges. I calculated the availability of cutblocks and roads to an individual animal as the average distance to cutblock and road at ‘available’ locations sampled in the home range of that animal (see Matthiopoulos et al. 2011).

Clearly, other habitat characteristics besides forestry disturbance influence caribou distribution (e.g., terrain and vegetation, as they relate to food and predation risk; see for example, Terry et al. 2000, Apps et al. 2001, Johnson et al. 2004, Apps & McLellan 2006), and I initially attempted to fit models with some of these other factors. However, this led to the development of complex habitat models that did not improve my ability to estimate the effects of disturbance on caribou distribution. Instead, model building and interpretation became very challenging due to the difficulty in fitting a generalizable model to unique habitats across a DU. However, I did evaluate correlation between the forestry disturbance factors modeled here and other habitat features measured using spatial data (e.g., elevation, slope, forest stand characteristics) to ensure they were not completely confounded. Specifically, I tested for correlations using a Spearman correlation, ensuring correlation coefficients between covariates were less than 0.7, and by evaluating variance inflation factors (VIFs) in generalized linear models (GLMs), and ensuring VIFs were less than 10 ( DeCesare et al. 2012). This provided reassurance that distance to cublocks and roads were not confounded with each other or other habitat features potentially driving caribou distribution.

Results

DU6

In DU6, caribou location data was collected from 160 animals from the Calendar, Chinchaga, Maxhamish, Westside Fort Nelson (formerly seperated as Parker and Prophet), and Snake-Sahtaneh herds. The number of animals sampled in each herd ranged from 5 to 43, and the number of telemetry locations ranged from 8,144 to 57,307 per herd (Table 1).

Table 1. Number of Animals and Telemetry Locations Sampled by Caribou Herd in DU6 (Boreal).
Herd Animals Locations
Calendar 33 40937
Chinchaga 43 53993
Maxhamish 33 40295
Parker 7 8240
Prophet 5 8144
Snake-Sahtaneh 39 57307

Herd-Level Effect of Forestry on Caribou Distribution

There were some small differences in herd response to resource roads in DU6 (Fig. 2). For example, caribou in the Calendar, Chinchaga and Prophet herds tended to be further from roads then what was available on average in their home range. A random effect for distance to road was included in the GLMM to account for this variability.

There was variability in caribou response to cutblocks across herds in DU6 (Fig. 3). For example, caribou in the Calendar and Maxhamish were closer to cutblocks then what was available on average in their home range. A random effect for distance to cutblock was included in the GLMM to account for this variability.

Caribou-Forestry Disturbance Model

I fit a model with a functional response for distance to road but not cutblock (Table 2). There was a significant negative population-level (fixed) effect of roads on caribou, and clearly greater avoidance of roads by caribou as the average distance to road within a home range decreased (Fig. 4). Visual assessment of model predictions indicated no clear change in caribou use of habitat near cutblocks as the average distance to cublock in their home ranges changed. The model indicated an overall weak (non-significant), negative fixed effect of cutblocks on caribou distribution.

# model spec
model.lme4.du6.cut.rd.fxn <- glmer (pttype ~ dist_cut_min_all + 
                                                distance_to_resource_road +
                                                dist_rd_E +
                                                distance_to_resource_road*dist_rd_E +
                                      (dist_cut_min_all + distance_to_resource_road || HERD_NAME) +
                                      (dist_cut_min_all + distance_to_resource_road || animal_id),
                                      data = rsf.data.du6, 
                                      family = binomial (link = "logit"),
                                      verbose = T) 
# check residuals
binnedplot (fitted (model.lme4.du6.cut.rd.fxn), 
             residuals(model.lme4.du6.cut.rd.fxn, type = "response"), 
             nclass = NULL, 
             xlab = "Expected Values", 
             ylab = "Average residual", 
             main = "DU6 Model Binned Residual Plot", 
             cex.pts = 0.4, 
             col.pts = 1, 
             col.int = "red")
Table 2. Fixed Effect Coefficients of the DU6 Caribou-Forestry Disturbance Model.
Coefficient Coefficient.Estimate Std..Error z.value p.value
Intercept -1.451 0.170 -8.517 0.000
Distance to Cutblock 0.009 0.009 1.008 0.313
Distance to Resource Road 0.343 0.067 5.116 0.000
Average Distance to Resource Road (Home Range) -0.331 0.198 -1.670 0.095
Distance to Resource Road x Average Distance to Resource Road -0.262 0.090 -2.899 0.004

Figure 4. Functional response between distance to road and average distance to road in a caribou home range in DU6.

Figure 4. Functional response between distance to road and average distance to road in a caribou home range in DU6.

Random effects in the DU6 model indicated that caribou response to roads was relatively consistent across herds (Table 3). However, there was much more variability in caribou herd-level response to cutblocks. Caribou in the Calendar, Chinchaga and Prophet herds showed weak use of habitat near cutblocks, whereas caribou in the Maxhamish, Parker and Snake-Sahtaneh showed avoidance of cutblocks.

Table 3. Random Effect Herd-Level Coefficients of the DU6 Caribou-Forestry Disturbance Model.
Herd Intercept Distance.to.Cutblock Distance.to.Road
Calendar -1.111 -0.005 0.354
Chinchaga -1.613 -0.002 0.375
Maxhamish -1.526 0.016 0.335
Parker -1.517 0.036 0.346
Prophet -1.283 -0.001 0.335
Snake-Sahtaneh -1.654 0.009 0.316

DU7

In DU7, caribou location data was collected from 131 animals from the Charlotte Alplands, Chase, Finlay, Frog, Graham, Itch-Ilgachuz, Muskwa, Pink Mountain, Rainbows, Spatsizi, Telkwa, Tsenaglode and Tweedsmuir herds. The number of animals sampled ranged from 1 to 35 per herd and the number of telemetry locations ranged from 856 to 41,837 per herd (Table 4).

Table 4. Number of Animals and Telemetry Locations Sampled by Caribou Herd in DU7.
Herd Animals Locations
Charlotte Alplands 4 3908
Chase 4 5827
Finlay 1 3020
Frog 4 8783
Graham 17 20473
Itcha-Ilgachuz 21 41837
Muskwa 19 10123
Pink Mountain 8 2599
Rainbows 9 8249
Spatsizi 5 12232
Telkwa 1 856
Tsenaglode 3 1381
Tweedsmuir 35 32131

Herd-Level Effect of Forestry on Caribou Distribution

There was variability in caribou response to resource roads across caribou herds in DU7 (Fig. 5). For example, caribou in the Chase, Graham, Itcha-Ilgachuz, Tsenaglode and Tweedsmuir tended to be further from roads, whereas caribou in the Frog tended ot be closer to roads then what was available on average in their home ranges. A random effect for distance to road was included in the GLMM to account for this variability.

There was variability in caribou response to cutblocks across herds in DU7 (Fig. 6). For example, caribou in the Itcha-Ilgachuz, Muskwa, Pink Mountain and Rainbows herds used areas further from cutblocks, whereas cariou in the Frog and Tsenaglode herds used areas closer to cutblocks then what was available on average in their home range. A random effect for distance to cutblock was included in the GLMM to account for this variability.

Caribou-Forestry Disturbance Model

I fit a model with a functional response for distance to road but not cutblock (Table 5). There was a significant negative fixed effect of roads on caribou distribution, and there was greater avoidance of roads by caribou as the distance to roads within home ranges decreased (Fig. 7). The model indicated a weak (non-significant), negative fixed effect of cutblocks on caribou distribution. Visual assessment of model predictions indicated no clear change in caribou use of habitat near cutblocks as the average distance to cublock in their home ranges changed.

# model spec
model.lme4.du7.cut.rd.fxn <- glmer (pttype ~ dist_cut_min_all +
                                             distance_to_resource_road +
                                             dist_rd_E +
                                             distance_to_resource_road*dist_rd_E +
                                        (dist_cut_min_all + distance_to_resource_road || animal_id) +                                         (dist_cut_min_all + distance_to_resource_road || HERD_NAME),
                                  data = rsf.data.du7,
                                  family = binomial (link = "logit"),
                                  verbose = T)
# check residuals
binnedplot (fitted (model.lme4.du7.cut.rd.fxn), 
             residuals(model.lme4.du7.cut.rd.fxn, type = "response"), 
             nclass = NULL, 
             xlab = "Expected Values", 
             ylab = "Average residual", 
             main = "DU7 Model Binned Residual Plot", 
             cex.pts = 0.4, 
             col.pts = 1, 
             col.int = "red")
Table 5. Fixed Effect Coefficients of the DU7 Caribou-Forestry Disturbance Model.
Coefficient Coefficient.Estimate Std..Error z.value p.value
Intercept -1.373 0.218 -6.293 0.000
Distance to Cutblock 0.003 0.010 0.329 0.742
Distance to Resource Road 0.081 0.025 3.220 0.001
Average Distance to Resource Road (Home Range) -0.016 0.020 -0.813 0.416
Distance to Resource Road x Average Distance to Resource Road -0.003 0.002 -1.440 0.150

Figure 7. Functional response between distance to road and average distance to road in a caribou home range in DU7.

Figure 7. Functional response between distance to road and average distance to road in a caribou home range in DU7.

Random effects in the DU7 model indicated that caribou avoided roads consistently across herds, although there was significant variability in the strength of avoidance (Table 6). There was more variability in caribou response to cutblocks. Caribou in the Charlotte Alplands, Finlay, Frog, Graham, Pink Mountain, Spatsizi, Tsenaglode and TWeedsmuir tended to be closer to cutblocks, whereas caribou in other herds tended to be further from cutblocks then what was available in their home ranges.

Table 6. Random Effect Herd-Level Coefficients of the DU7 Caribou-Forestry Disturbance Model.
Herd Intercept Distance.to.Cutblock Distance.to.Road
Charlotte Alplands -0.758 -0.040 0.207
Chase -1.305 0.003 0.075
Finlay -0.934 -0.015 0.051
Frog -0.673 -0.008 0.085
Graham -1.801 -0.011 0.147
Itcha-Ilgachuz -1.041 0.002 0.018
Muskwa -2.806 0.017 0.042
Pink Mountain -2.174 -0.009 0.147
Rainbows -0.656 0.064 0.024
Spatsizi -1.379 -0.002 0.057
Telkwa -1.751 0.043 0.097
Tsenaglode -0.927 -0.001 0.083
Tweedsmuir -1.632 -0.002 0.027

DU8

In DU8, caribou location data was collected from 152 animals from the Burnt Pine, Kennedy Siding, Moberly, Narraway, Quintette and Scott herds. The number of animals sampled per herd ranged from 8 to 35 and the number of telemetry locations per herd ranged from 5,414 to 65,431 per herd (Table 7).

Table 7. Number of Animals and Telemetry Locations Sampled by Caribou Herd in DU8.
Herd Animals Locations
Burnt Pine 16 5414
Kennedy Siding 38 31502
Moberly 14 13946
Narraway 19 11558
Quintette 57 65431
Scott 8 10164

Herd-Level Effect of Forestry on Caribou Distribution

There was variability in caribou response to resource roads across caribou herds in DU8 (Fig. 8). For example, caribou in the Burnt Pine, Kennedy Siding and Scott herds tended to be closer to roads, whereas caribou in the Moberly, Narraway and Quintette tended to be further from roads then what was available. A random effect for distance to road was included in the GLMM to account for this variability.

There was variability in caribou response to cutblocks across herds in DU8 (Fig. 9). For example, caribou in the Quintette herd tended to use areas further from cutblocks, whereas caribou in the Narraway herd tended to be closer to cutblocks then what was available. A random effect for distance to cutblock was included in the GLMM to account for this variability.

Caribou-Forestry Disturbance Model

I fit a model with a functional response for distance to road but not cutblock (Table 8). The model indicated a positive population-level (fixed) effect of cutblocks on caribou distribution and negative population-level effect of roads on caribou distribution. Visual assessment of model predictions indicated no clear change in caribou use of habitat near cutblocks as the average distance to cublock in their home ranges changed. However, there was greater avoidance of roads by caribou as the average distance to road within home ranges decreased (Fig. 10).

# model spec
model.lme4.du8.cut.rd.fxn <- glmer (pttype ~ dist_cut_min_all + 
                                            distance_to_resource_road +
                                            dist_rd_E + 
                                            distance_to_resource_road*dist_rd_E +
                                  (dist_cut_min_all + distance_to_resource_road || HERD_NAME) +
                                  (dist_cut_min_all + distance_to_resource_road || animal_id),
                                  data = rsf.data.du8, 
                                  family = binomial (link = "logit"),
                                  verbose = T) 
# check residuals
binnedplot (fitted (model.lme4.du8.cut.rd.fxn), 
             residuals(model.lme4.du8.cut.rd.fxn, type = "response"), 
             nclass = NULL, 
             xlab = "Expected Values", 
             ylab = "Average residual", 
             main = "DU8 Model Binned Residual Plot", 
             cex.pts = 0.4, 
             col.pts = 1, 
             col.int = "red")
Table 8. Fixed Effect Coefficients of the DU8 Caribou-Forestry Disturbance Model.
Coefficient Coefficient.Estimate Std..Error z.value p.value
Intercept -1.694 0.152 -11.116 0.000
Distance to Cutblock -0.079 0.031 -2.559 0.010
Distance to Resource Road 0.239 0.067 3.580 0.000
Average Distance to Resource Road (Home Range) -0.018 0.063 -0.281 0.778
Distance to Resource Road x Average Distance to Resource Road -0.056 0.026 -2.174 0.030

Figure 10. Functional response between distance to road and average distance to road in a caribou home range in DU8.

Figure 10. Functional response between distance to road and average distance to road in a caribou home range in DU8.

Random effects in the DU8 model indicated that caribou avoided roads consistently across herds, although there was variability in the strength of avoidance (Table 9). Caribou herds were also generally closer to cutblocks then what was available, with the exception of the Quintette herd, where caribou were further from cutblocks then what was available.

Table 9. Random Effect Herd-Level Coefficients of the DU8 Caribou-Forestry Disturbance Model.
Herd Intercept Distance.to.Cutblock Distance.to.Road
Burnt Pine -1.807 -0.036 0.132
Kennedy Siding -1.720 -0.170 0.351
Moberly -1.810 -0.148 0.304
Narraway -1.819 -0.066 0.291
Quintette -1.758 0.012 0.193
Scott -1.244 -0.063 0.163

DU9

In DU9, caribou location data was collected from 25 animals from the Hart Ranges, Nakusp and South Selkirks herds (Table 10). The number of animals sampled per herd ranged from 6 to 11 and the number of telemetry locations per herd ranged from 1,821 to 4,917 per herd.

Table 10. Number of Animals and Telemetry Locations Sampled by Caribou Herd in DU9.
Herd Animals Locations
Hart Ranges 11 4917
Nakusp 8 1821
South Selkirks 6 2629

Herd-Level Effect of Forestry on Caribou Distribution

There was variability in caribou response to resource roads across caribou herds in DU9 (Fig. 11). For example, caribou in the Hart Ranges tended to be closer to roads, whereas caribou in the Nakusp and South Slekirks tended to be further from roads then what was available. A random effect for distance to road was included in the GLMM to account for this variability.

There was variability in caribou response to cutblocks across herds in DU9 (Fig. 12). For example, caribou in the Hart Ranges herd tended to use areas closer to cutblocks, whereas caribou in the Nakusp herd tended to be further from cutblocks then what was available. A random effect for distance to cutblock was included in the GLMM to account for this variability.

Caribou-Forestry Disturbance Model

I fit a model with a functional response for distance to road but not cutblock (Table 11). The model included negative population-level (fixed) effects of cutblocks and roads on caribou distribution. However, the effect of cutblocks was weaker than for roads. Visual assessment of model predictions indicated no clear change in caribou use of habitat near cutblocks as the average distance to cublock in their home ranges changed. However, there was greater avoidance of roads by caribou as the distance to roads within home ranges decreased (Fig. 13).

# model spec
model.lme4.du9.cut.rd.fxn <- glmer (pttype ~ dist_cut_min_all + 
                                                distance_to_resource_road +
                                                dist_rd_E +
                                                distance_to_resource_road*dist_rd_E +
                                      (dist_cut_min_all + distance_to_resource_road || HERD_NAME) +
                                      (dist_cut_min_all + distance_to_resource_road || animal_id),
                                      data = rsf.data.du9, 
                                      family = binomial (link = "logit"),
                                      verbose = T) 
# check residuals
binnedplot (fitted (model.lme4.du9.cut.rd.fxn), 
             residuals(model.lme4.du9.cut.rd.fxn, type = "response"), 
             nclass = NULL, 
             xlab = "Expected Values", 
             ylab = "Average residual", 
             main = "DU9 Model Binned Residual Plot", 
             cex.pts = 0.4, 
             col.pts = 1, 
             col.int = "red")
Table 11. Fixed Effect Coefficients of the DU9 Caribou-Forestry Disturbance Model.
Coefficient Coefficient.Estimate Std..Error z.value p.value
Intercept -1.979 0.499 -3.964 0.000
Distance to Cutblock 0.043 0.070 0.617 0.537
Distance to Resource Road 0.118 0.088 1.341 0.180
Average Distance to Resource Road (Home Range) -0.165 0.093 -1.787 0.074
Distance to Resource Road x Average Distance to Resource Road -0.006 0.016 -0.375 0.708

Figure 13. Functional response between distance to road and average distance to road in a caribou home range in DU9.

Figure 13. Functional response between distance to road and average distance to road in a caribou home range in DU9.

Random effects in the DU9 model indicated that caribou avoided roads consistently across herds, although there was variability in the strength of avoidance (Table 12). There was variability in use of areas close to cutblocks across herds, where caribou in the Nakusp herd tended to be further from cutblocks then what was available, compared to other herds.

Table 12. Random Effect Herd-Level Coefficients of the DU9 Caribou-Forestry Disturbance Model.
Herd Intercept Distance.to.Cutblock Distance.to.Road
Hart Ranges -0.963 -0.024 0.083
Nakusp -2.424 0.168 0.007
South Selkirks -2.539 -0.011 0.262

Conclusions

Research has repeatedly shown that caribou are negatively influenced by forestry activity. This can include negative density effects, for example, on caribou survival ( Wittmer et al. 2007, Johnson et al. 2015), and behavioural effects, for example, on caribou distribution (e.g., Courbin et al. 2009, DeCesare et al. 2012, Mumma et al. 2018). Here we built on this evidence and developed a CFDM that empirically related forestry disturbance to caribou distribution. The model provides an indicator of forestry influences on caribou habitat, and thus can be linked to forestry simulator models, i.e., the caribou and land use simualtor (CLUS), to indicate spatial-temporal changes in caribou habitat quality from forestry.

Our CFDM showed a strong negative effect of resource roads on caribou distribution. Roads are likely representative of many ecological effects of resource development on caribou. Roads can act as semi-permeable barriers to caribou, resulting in loss of functional habitat due to caribou avoidance of areas near roads ( Dyer et al. 2001, Dyer et al. 2002, Wilson et al. 2016). Caribou may also avoid roads to minimize predation risk ( Dussault et al. 2012), as predators such as wolves and bears may make use of these areas to hunt prey ( Whittington et al. 2011, Latham et al 2011). The CFDM does not attempt to measure or disentangle the effects of these different ecological processes, but it is assumed that the negative relationship between caribou and roads is indicative of these processes.

We identified a functional response between roads and caribou distribution, and included this in our model. Functional responses show how animal use of habitat changes as a function of the availability of that habitat to them on the landscape. Functional responses can be particularly important for understanding and modeling the effects of disturbances such as forestry on caribou, as caribou avoidance, or use, of disturbance features likely changes as the landscape is increasingly disturbed. We found that caribou were more likely to avoid roads in landscapes with more roads. This suggests that habitat patches far from roads become increasingly important to caribou as the amount of roads in their home ranges increases. This effect is important to consider when simulating the effects of forestry disturbance on caribou through space and time. The development of roads in landscapes with few existing roads will have a relatively small effect on caribou use compared to landscapes where more road development has already occurred.

In general, the CFDM indicated that the effect of cutblocks on caribou distribution was negative, but more variable and weaker than the effect of roads. Specifically, while the population-level (fixed) effects of cutblocks was negative in DU’s 6, 7 and 9, it was positive in DU 8, and there was more variability in the herd-level effects. Furthermore, there was no clear functional response in use of areas near cutblocks by caribou as the amount of cutblocks in caribou home ranges changed. Cutblocks can have a negative effect on caribou, as they provide more forage for ungulates such as moose, thus supporting higher densities of ungulates and their predators, such as wolves, that then predate on caribou ( Seip 1992, DeCesare et al. 2010, Serrouya et al. 2017). However, there may be reasons why we detected a weak negative, and sometimes positive effect of cutblocks on caribou. First, caribou may use cutblocks periodically throughout the year to access forage. Caribou will select for deciduous shrub species during summer foraging ( Denryter et al. 2017), and these species tend to increase in cutblocks ( Strong and Gates 2006). Resarch in Quebec found that caribou selected younger cutblocks, presumably because of their forage benefit ( Dussault et al. 2012). Second, the negative effect of cutblocks may occur at larger scales, outside of the home range. DeCesare et al. 2012 found that caribou avoided cutblocks at the range scale (i.e., caribou home ranges occurred in areas with fewer cutblocks), but not within their home ranges. This suggests the negative predation effect of cutblocks occurs at larger scales, that caribou are less likely to establish home ranges in areas with more cutblocks, perhaps because they are avoiding predation risk or they are unable to survive in those areas. Clearly, the effects of cutblocks on caribou distribution is more nuanced than roads, and the CFDM appears to represent these nuanced effects by representing a weak negative effect of cutblocks on caribou in most DUs, that varies at the herd-level.

The CFDM provides a framework for testing the effects of simulated future forestry activities on caribou. It includes random effects to account for individual and herd-level variability in caribou responses to cutblocks and roads, and a functional response to account for the effect of landscape-level disturbances on local use of habitat. These features provide a great deal of flexibility in estimating the effects of proposed future forestry acitivities on caribou.